Method and device for obtaining equivalent relation between FRP laminated plates
The deflection and thickness prediction model was trained through the artificial neural network method, and a series neural network model was established based on the principle of maximum deflection equivalent, which solved the problem of difficulty in obtaining the thickness equivalent relationship between FRP laminates, and achieved rapid and effective equivalent relationship prediction, reducing the experimental cost and design cycle.
Patent Information
- Application Number
- CN202411483005.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-23
- Publication Date
- 2025-03-04
AI Technical Summary
The prior art lacks an effective method to quickly obtain the thickness equivalent relationship between FRP laminates under the action of far-field explosion loads, making it difficult to establish a thickness equivalent model between FRP laminates.
Using the artificial neural network method, the deflection prediction and thickness prediction neural network models are trained, and the tandem neural network model is established based on the principle of maximum deflection equivalent, and the thickness equivalent relationship between FRP laminates is quickly obtained.
This realizes direct prediction of the complex mechanism of action between the explosion load and the FRP laminate without considering the need to consider, simplifying the acquisition of the equivalent relationship between the FRP laminate and reducing the test cost and design cycle.
Smart Images
Figure CN119560064A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of fiber reinforced composite materials (FRP), and in particular relates to a method and a device for obtaining an equivalent relationship between FRP laminates. Background Art
[0002] Fiber reinforced composite (FRP) laminates are widely used in the lightweight design of special equipment such as aircraft, ships, and special vehicles due to their excellent properties such as high specific strength and high specific modulus. In certain application scenarios, these special equipment will face the threat of explosion impact, which puts forward higher requirements on the explosion resistance of composite materials. With the continuous improvement of the comprehensive performance of new composite materials, explosion-proof structures are driven towards lightweight development, but the cost of use has also soared.
[0003] Obtaining the equivalent relationship between new high-performance composite materials and conventional low-cost composite materials makes it possible to use low-cost composite materials instead of new composite materials in the research stage, which helps to reduce the cost of large-scale structural explosion-resistant tests and shorten the test cycle. However, there is currently a lack of effective methods to quickly obtain the thickness equivalence relationship between FRP laminates with different mechanical properties under explosion loads. The main difficulties are: first, the explosion loading conditions are very dangerous, and there are few units that have the conditions to carry out relevant tests. Secondly, as an anisotropic material, FRP laminates have many related mechanical performance parameters and complex energy absorption mechanisms, making it difficult to establish relevant numerical analysis equivalent models. Existing studies are all based on a simple comparison of the explosion response process of several different FRP laminates based on a small number of experiments. There is no systematic study on the equivalence between FRP laminates with different mechanical properties, which makes it difficult to directly establish a thickness equivalence model between FRP laminates through limited experiments or numerical simulation methods. Summary of the invention
[0004] The present invention provides a method and device for obtaining an equivalent relationship between FRP laminates, which are used to solve the technical problem of how to quickly obtain the thickness equivalent relationship between FRP laminates under the action of a far-field explosion, and help to reduce the design cycle and test cost of composite explosion-proof structures.
[0005] In a first aspect, a method for obtaining an equivalent relationship between FRP laminates is disclosed, the method comprising the following steps:
[0006] Step S1: taking experimental data of various types of FRP laminates as sample data; selecting training data from the sample data; extracting the FRP laminate thickness TA, geometric features, material properties and explosive loading conditions of the first type of FRP laminate to be equivalent to the first type of FRP laminate from the training data, and constructing a first feature vector; extracting the maximum deflection, geometric features, material properties and explosive loading conditions of the second type of FRP laminate to be equivalent to the first type of FRP laminate from the training data, and constructing a second feature vector;
[0007] Step S2: training a deflection prediction neural network model based on the first eigenvector to obtain a trained deflection prediction neural network model; the deflection prediction neural network model takes the first eigenvector as input and determines the maximum deflection W of the first type FRP laminate during the explosion response based on the first eigenvector;
[0008] Step S3: training a thickness prediction neural network model based on the second eigenvector to obtain a trained thickness prediction neural network model; the thickness prediction neural network model takes the second eigenvector as input to generate a thickness TB of the second type FRP laminate corresponding to the maximum deflection of the second type FRP laminate;
[0009] Step S4: sequentially connecting the trained deflection prediction neural network model and the trained thickness prediction neural network model in series to form a trained series neural network model;
[0010] Step S5: Obtain the FRP laminate thickness T of the first FRP laminate for which the equivalent relationship is to be obtained A1 , geometric features, material properties and explosive loading conditions, construct a third eigenvector; obtain the geometric features, material properties and explosive loading conditions of the second FRP laminate whose equivalent relationship is to be obtained; the trained series neural network model generates the maximum deflection W corresponding to the first FRP laminate based on the third eigenvector A1 , based on the maximum deflection W A1 , constructing a fourth eigenvector based on the geometric features, material properties and explosion loading conditions of the second FRP laminate to obtain the equivalent relationship, and generating the thickness of the second FRP laminate equivalent to the first FRP laminate based on the fourth eigenvector.
[0011] Preferably, the model structures of the deflection prediction neural network model and the thickness prediction neural network model both include an input layer, a first hidden layer, a second hidden layer and an output layer connected in sequence.
[0012] Preferably, the activation functions used in the first hidden layer and the second hidden layer of the deflection prediction neural network model and the thickness prediction neural network model are
[0013]
[0014] Among them, x is the input value of the activation function, f(x) is the output value of the activation function, α is a hyperparameter, and the value range of α is 0 to 1.
[0015] Preferably, the activation function used in the output layer is:
[0016]
[0017] Among them, f2(x1) is the output value of the activation function of the output layer, x1 is the input value of the activation function used by the output layer, and β is a hyperparameter.
[0018] Preferably, the loss functions of the deflection prediction neural network model and the thickness prediction neural network model are both:
[0019]
[0020] Among them, Loss log-cosh is the prediction error, is the i1th true value, is the i1th prediction value, N is the number of training samples, i1 is the sample number, and cosh is the hyperbolic cosine function.
[0021] Preferably, the parameter update formulas of the deflection prediction neural network model and the thickness prediction neural network model are both:
[0022]
[0023] Among them, Δω ij ω ij The correction amount, η is the learning rate;
[0024]
[0025] Among them, the number of iterations of training the deflection prediction neural network model and the thickness prediction neural network model are both n.
[0026] In a second aspect, a device for obtaining an equivalent relationship between FRP laminates is disclosed, the device comprising:
[0027] A feature construction module: configured to use experimental data of multiple types of FRP laminates as sample data; select training data from the sample data; extract the FRP laminate thickness TA, geometric features, material properties and explosive loading conditions of the first type of FRP laminate to be equivalent to the first type of FRP laminate from the training data to construct a first feature vector; extract the maximum deflection, geometric features, material properties and explosive loading conditions of the second type of FRP laminate to be equivalent to the first type of FRP laminate from the training data to construct a second feature vector;
[0028] A first training module: configured to train a deflection prediction neural network model based on the first eigenvector to obtain a trained deflection prediction neural network model; the deflection prediction neural network model takes the first eigenvector as input and determines the maximum deflection W of the first type of FRP laminate during the explosion response process based on the first eigenvector;
[0029] A second training module: configured to train a thickness prediction neural network model based on the second eigenvector to obtain a trained thickness prediction neural network model; the thickness prediction neural network model uses the second eigenvector as input to generate a thickness TB of the second type FRP laminate corresponding to the maximum deflection of the second type FRP laminate;
[0030] A series connection module: configured to sequentially connect the trained deflection prediction neural network model and the trained thickness prediction neural network model in series to serve as a trained series connection neural network model;
[0031] Equivalent module: configured to obtain the FRP laminate thickness T of the first FRP laminate to be obtained for the equivalent relationship A1 , geometric features, material properties and explosive loading conditions, construct a third eigenvector; obtain the geometric features, material properties and explosive loading conditions of the second FRP laminate whose equivalent relationship is to be obtained; the trained series neural network model generates the maximum deflection W corresponding to the first FRP laminate based on the third eigenvector A1 , based on the maximum deflection W A1 , constructing a fourth eigenvector based on the geometric features, material properties and explosion loading conditions of the second FRP laminate to obtain the equivalent relationship, and generating the thickness of the second FRP laminate equivalent to the first FRP laminate based on the fourth eigenvector.
[0032] In a third aspect, an electronic device is disclosed, the electronic device comprising:
[0033] at least one processor; and
[0034] a memory communicatively connected to the at least one processor; wherein,
[0035] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described above.
[0036] In a fourth aspect, a non-transitory computer-readable storage medium storing computer instructions is disclosed, wherein the computer instructions are used to cause the computer to execute the method as described above.
[0037] Artificial neural network (ANN) models can discover patterns and laws by learning the complex relationship between input and output, and have excellent performance in processing nonlinear, high-dimensional and complex relationship data. In order to solve the existing problems, the present invention introduces the ANN method and establishes a series neural network model based on the maximum deflection equivalence principle to quickly obtain the thickness equivalence relationship between FRP laminates under far-field explosion.
[0038] The present invention has the following technical effects:
[0039] 1. The present invention adopts an artificial neural network method, which directly predicts the explosion response results of FRP laminates without considering the complex action mechanism and influencing factors between the explosion load and the FRP laminates.
[0040] 2. The present invention connects the deflection prediction and thickness prediction neural network models in series to form a series neural network model. Only by inputting the relevant parameters of the two FRP laminates and the explosion loading conditions, the thickness equivalence relationship between the two can be quickly obtained.
[0041] 3. The present invention can adjust the input neuron variables, quantity and value range of the neural network model according to demand, and study the influence of different parameters on the thickness equivalence relationship between FRP laminates, and has a wide range of applications.
[0042] 4. The present invention only needs to carry out corresponding experiments or simulation calculations for different input neurons, and the obtained data set can be used for deflection prediction and thickness prediction neural network model training at the same time, greatly reducing the workload and experimental costs.
[0043] 5. After training, the series neural network model proposed in the present invention can directly predict the thickness equivalence relationship between FRP laminates with any value within the input variable range, and has strong practicality. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] Figure 1 A schematic diagram of a method for obtaining an equivalent relationship between FRP laminates;
[0045] Figure 2 Schematic diagram of the architecture of the method for obtaining the equivalent relationship between FRP laminates;
[0046] Figure 3 It is the structural schematic diagram of the deflection prediction model;
[0047] Figure 4 It is a schematic diagram of the thickness prediction model structure;
[0048] Figure 5 Schematic diagram of input neurons for the output layer;
[0049] Figure 6 Schematic diagram of the structure of the device for obtaining the equivalent relationship between FRP laminates. DETAILED DESCRIPTION
[0050] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.
[0051] like Figure 1-Figure 2 As shown, the present invention provides a method for obtaining an equivalent relationship between FRP laminates, the method comprising the following steps:
[0052] Step S1: taking experimental data of various types of FRP laminates as sample data; selecting training data from the sample data; extracting the FRP laminate thickness T of the first type of FRP laminate for which an equivalent relationship is to be obtained from the training data A , geometric features, material properties and explosive loading conditions to construct a first feature vector; extracting the maximum deflection, geometric features, material properties and explosive loading conditions of the second type of FRP laminate to be determined to have an equivalent relationship with the first type of FRP laminate from the training data to construct a second feature vector;
[0053] Step S2: training a deflection prediction neural network model based on the first eigenvector to obtain a trained deflection prediction neural network model; the deflection prediction neural network model takes the first eigenvector as input and determines the maximum deflection W of the first type of FRP laminate during the explosion response process based on the first eigenvector;
[0054] Step S3: training a thickness prediction neural network model based on the second eigenvector to obtain a trained thickness prediction neural network model; the thickness prediction neural network model uses the second eigenvector as input to generate a thickness T of the second type FRP laminate corresponding to the maximum deflection of the second type FRP laminate. B ;
[0055] Step S4: sequentially connecting the trained deflection prediction neural network model and the trained thickness prediction neural network model in series to form a trained series neural network model;
[0056] Step S5: Obtain the FRP laminate thickness T of the first FRP laminate for which the equivalent relationship is to be obtained A1, geometric features, material properties and explosive loading conditions, construct a third eigenvector; obtain the geometric features, material properties and explosive loading conditions of the second FRP laminate whose equivalent relationship is to be obtained; the trained series neural network model generates the maximum deflection W corresponding to the first FRP laminate based on the third eigenvector A1 , based on the maximum deflection W A1 , constructing a fourth eigenvector based on the geometric features, material properties and explosion loading conditions of the second FRP laminate to obtain the equivalent relationship, and generating the thickness of the second FRP laminate equivalent to the first FRP laminate based on the fourth eigenvector.
[0057] Among them, the model structures of the deflection prediction neural network model and the thickness prediction neural network model both include an input layer, a first hidden layer, a second hidden layer and an output layer connected in sequence.
[0058] In the present invention, the trained deflection prediction neural network model and the trained thickness prediction neural network model are sequentially connected in series based on the maximum deflection equivalence principle. The maximum equivalence principle is the maximum deflection equivalence principle, which is a method for structural analysis and design, and is mainly used to evaluate and optimize the performance of engineering structures under external loads. The core idea of this principle is to establish an equivalent relationship by comparing the deflections (i.e., the degree of deformation of the structure under load) generated by different materials, shapes or structural designs, thereby simplifying the analysis process of complex structures. It is often used in thin-walled structure analysis and explosion-proof performance evaluation.
[0059] like Figure 3-Figure 5 As shown, the input neurons of the deflection prediction neural network model include variables such as geometric characteristics, material properties and explosive loading conditions of the FRP laminate. The input layer of the deflection prediction neural network model consists of several input neurons, among which the thickness of the FRP laminate is a fixed input neuron, and other input neurons can be adjusted according to actual needs.
[0060] The hidden layer of the deflection prediction neural network model is composed of a number of hidden layer neurons, and each hidden layer neuron is connected to each input and output neuron through an independent weight and threshold.
[0061] The output layer of the deflection prediction neural network model includes a fixed output neuron: the maximum deflection W of the FRP laminate explosion response process.
[0062] The thickness prediction neural network model is used to predict the thickness of the FRP laminate when the maximum deflection in the explosion response process is known, and comprises an input layer, a hidden layer and an output layer.
[0063] The input layer of the thickness prediction neural network model is composed of a plurality of input neurons, wherein the maximum deflection during the explosion response of the FRP laminate is a fixed input neuron, and other input neurons can be adjusted according to actual needs.
[0064] The input neurons of the thickness prediction neural network model are the same as the input neurons in the deflection prediction neural network model except that they additionally include the maximum deflection W of the FRP laminate, and will not be repeated here.
[0065] The hidden layer of the thickness prediction neural network model is composed of a number of hidden layer neurons, and its structure is the same as that of the hidden layer of the deflection prediction neural network model, which will not be repeated here.
[0066] The output layer of the thickness prediction neural network model includes a fixed output neuron: the thickness T of the FRP laminate.
[0067] The neural network input neuron-related variables can be selected according to research needs, or variables with less impact can be eliminated through sensitivity analysis to reduce the complexity of the neural network model and the size of the data set used for training.
[0068] The series neural network model is composed of a trained deflection prediction neural network model and a thickness prediction neural network model connected in series. Therefore, the input neuron of the series neural network model is the thickness T of the FRP laminate A. A As well as other geometric characteristics, material properties and explosion loading conditions of FRP laminate A and FRP laminate B; the output neuron is the thickness T of FRP laminate B B .
[0069] The basic working principle of the series neural network model is: based on the maximum deflection equivalence principle, the maximum deflection prediction result of the FRP laminate A predicted by the deflection prediction neural network model is used as the input neuron of the thickness prediction neural network model to predict the thickness of the FRP laminate B, that is, the thickness T is obtained. A The FRP laminate A with a thickness of T B FRP laminate B is equivalent.
[0070] Furthermore, the geometric features include the in-plane dimensions (side length or diameter), thickness, number and size of the reinforcement strips of the FRP laminate; the material properties include the density ρ, longitudinal elastic modulus E11, longitudinal tensile strength XT, in-plane shear modulus G12, etc. of the FRP laminate; the explosion loading conditions can be the maximum overpressure, explosion impulse, explosive equivalent and detonation distance, etc. In addition, the present invention establishes an equivalent model based on the maximum deflection equivalent principle, and the explosion loading conditions must meet the far-field explosion conditions to ensure that the FRP laminate has large deformation as the main failure mode without causing major damage.
[0071] like Figure 5 As shown, TA is the thickness of FRP laminate A, W is the maximum deflection, T B is the deflection of FRP laminate B, T-FRP is the thickness of the laminate, E 11 is the longitudinal elastic modulus, ρ is the density, XT is the longitudinal tensile strength, G 12 is the in-plane shear modulus.
[0072] After confirming the input neurons, relevant experiments and simulations were carried out to obtain the explosion response results of FRP laminates under different variable values, and used as data sets to train the deflection prediction and thickness prediction neural network models respectively.
[0073] The data set must ensure that each input neuron has no less than 5 values and performs full permutation and combination. Among them, the thickness T of the FRP laminate in the data set used to train the deflection prediction neural network model is one of the input quantities, and the maximum deflection W is the output quantity; the maximum deflection W of the FRP laminate in the data set used to train the thickness prediction neural network model is one of the input quantities, and the thickness T is the output quantity.
[0074] The output z of the j1th neuron in the first hidden layer of the deflection prediction neural network model and the thickness prediction neural network model is j1 It is obtained from the following formula:
[0075]
[0076] Among them, x i is the i-th input neuron in the input layer, For x i With z j1 The weight between Yes j1 The threshold is used to control the activation function f(·), k is the number of neurons in the input layer, i is the input neuron number, n1 is the number of neurons in the first hidden layer, j1 is the neuron number in the first hidden layer, is the output of the j1th neuron in the first hidden layer.
[0077] The output z of the j2th neuron in the second hidden layer is j2 It is obtained from the following formula:
[0078]
[0079] in, For z j1 With z j2 The weight between Yes j2 The threshold value, n2 is the number of neurons in the second hidden layer, j2 is the neuron number in the second hidden layer, is the output of the j1th neuron in the first hidden layer, is the output of the j2th neuron in the second hidden layer.
[0080]
[0081] in, For z j With z j2 The weight between j Yes j The threshold is used to control the activation function f2(·), n is the number of neurons in the output layer, j is the number of neurons in the output layer, n2 is the number of neurons in the second hidden layer, j2 is the number of neurons in the second hidden layer, z j is the output of the jth output neuron in the output layer, is the output of the j2th neuron in the second hidden layer.
[0082] Furthermore, the activation functions used in the first hidden layer and the second hidden layer of the deflection prediction neural network model and the thickness prediction neural network model are
[0083]
[0084] Among them, x is the input value of the activation function, f(x) is the output value of the activation function, α is a hyperparameter, and the value range of α is 0 to 1.
[0085] In this embodiment, when the thickness of the laminate is small, a value of 0.01*T is recommended.
[0086] The activation function used in the output layer is:
[0087]
[0088] Among them, f2(x1) is the output value of the activation function of the output layer, x1 is the input value of the activation function used by the output layer, and β is a hyperparameter.
[0089] β is usually set to 1, and can be increased appropriately if the prediction accuracy is not ideal.
[0090] The loss functions of the deflection prediction neural network model and the thickness prediction neural network model are both:
[0091]
[0092] Among them, Loss log-cosh is the prediction error, is the i1th true value, is the i1th prediction value, N is the number of training samples, i1 is the sample number, and cosh is the hyperbolic cosine function.
[0093] The parameter updating formulas of the deflection prediction neural network model and the thickness prediction neural network model are as follows:
[0094]
[0095] Among them, Δω ij ω ij The correction amount, η is the learning rate.
[0096]
[0097] Among them, the number of iterations of training the deflection prediction neural network model and the thickness prediction neural network model are both n.
[0098] In the present invention, the activation functions used in the hidden layer and the output layer in the neural network model are f(x) and f2(x1), respectively; and the Log-Cosh loss function is used to calculate the prediction error.
[0099] The deflection prediction neural network model and the thickness prediction neural network model can be trained using a common data set, and only the thickness variable and the maximum deflection variable in the data set need to be adjusted.
[0100] In the present invention, if the non-fixed input neurons in the deflection prediction and thickness prediction neural network models are the same, the two neural network models can share a data set for training, and only the thickness of the fixed neurons FRP laminate and the maximum deflection during the explosion response in the data set need to be swapped.
[0101] In the present invention, the sample data set is divided into a training set and a test set at a ratio of 9:1 for training and testing of the neural network model. After the training of the present invention is completed, the prediction accuracy of the deflection prediction and thickness prediction neural network models is tested, requiring that the prediction result error of the test set is less than 10%.
[0102] The present invention connects the trained deflection prediction and thickness prediction neural network models in series based on the maximum deflection equivalence principle to obtain a series neural network model for quickly obtaining the equivalent relationship between FRP laminates under far-field explosion conditions.
[0103] By simply inputting the explosion loading conditions, the geometric characteristics and material properties of the FRP laminate A and the FRP laminate B into the series neural network model, the thickness T of the FRP laminate A can be obtained. A With FRP laminate B thickness T B equivalence relationship.
[0104] like Figure 6 As shown, the present invention provides a device for obtaining an equivalent relationship between FRP laminates, the device comprising:
[0105] A feature construction module: configured to use experimental data of multiple types of FRP laminates as sample data; select training data from the sample data; extract the FRP laminate thickness TA, geometric features, material properties and explosive loading conditions of the first type of FRP laminate to be equivalent to the first type of FRP laminate from the training data to construct a first feature vector; extract the maximum deflection, geometric features, material properties and explosive loading conditions of the second type of FRP laminate to be equivalent to the first type of FRP laminate from the training data to construct a second feature vector;
[0106] A first training module: configured to train a deflection prediction neural network model based on the first eigenvector to obtain a trained deflection prediction neural network model; the deflection prediction neural network model takes the first eigenvector as input and determines the maximum deflection W of the first type of FRP laminate during the explosion response process based on the first eigenvector;
[0107] A second training module: configured to train a thickness prediction neural network model based on the second eigenvector to obtain a trained thickness prediction neural network model; the thickness prediction neural network model uses the second eigenvector as input to generate a thickness TB of the second type FRP laminate corresponding to the maximum deflection of the second type FRP laminate;
[0108] A series connection module: configured to sequentially connect the trained deflection prediction neural network model and the trained thickness prediction neural network model in series to serve as a trained series connection neural network model;
[0109] Equivalent module: configured to obtain the FRP laminate thickness T of the first FRP laminate to be obtained for the equivalent relationship A1 , geometric features, material properties and explosive loading conditions, construct a third eigenvector; obtain the geometric features, material properties and explosive loading conditions of the second FRP laminate whose equivalent relationship is to be obtained; the trained series neural network model generates the maximum deflection W corresponding to the first FRP laminate based on the third eigenvector A1 , based on the maximum deflection W A1 , constructing a fourth eigenvector based on the geometric features, material properties and explosion loading conditions of the second FRP laminate to obtain the equivalent relationship, and generating the thickness of the second FRP laminate equivalent to the first FRP laminate based on the fourth eigenvector.
[0110] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, it should be understood by those skilled in the art that the technical solutions described in the above embodiments can still be modified, or some or all of the technical features therein can be replaced by equivalents, and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for obtaining an equivalent relationship between FRP laminates, characterized in that: The method comprises the following steps: Step S1: taking experimental data of various types of FRP laminates as sample data; selecting training data from the sample data; extracting the FRP laminate thickness TA, geometric features, material properties and explosion loading conditions of the first type of FRP laminates for which an equivalent relationship is to be obtained from the training data, and constructing a first feature vector; Extracting the maximum deflection, geometric characteristics, material properties and explosion loading conditions of the second type of FRP laminate to be determined to have an equivalent relationship with the first type of FRP laminate from the training data, and constructing a second feature vector; Step S2: training a deflection prediction neural network model based on the first eigenvector to obtain a trained deflection prediction neural network model; the deflection prediction neural network model takes the first eigenvector as input and determines the maximum deflection W of the first type FRP laminate during the explosion response based on the first eigenvector; Step S3: training a thickness prediction neural network model based on the second eigenvector to obtain a trained thickness prediction neural network model; the thickness prediction neural network model takes the second eigenvector as input to generate a thickness TB of the second type FRP laminate corresponding to the maximum deflection of the second type FRP laminate; Step S4: sequentially connecting the trained deflection prediction neural network model and the trained thickness prediction neural network model in series to form a trained series neural network model; Step S5: Obtain the FRP laminate thickness T of the first FRP laminate for which the equivalent relationship is to be obtained A1 , geometric features, material properties and explosive loading conditions, construct a third eigenvector; obtain the geometric features, material properties and explosive loading conditions of the second FRP laminate whose equivalent relationship is to be obtained; the trained series neural network model generates the maximum deflection W corresponding to the first FRP laminate based on the third eigenvector A1 , based on the maximum deflection W A1 , constructing a fourth eigenvector based on the geometric features, material properties and explosion loading conditions of the second FRP laminate to obtain the equivalent relationship, and generating the thickness of the second FRP laminate equivalent to the first FRP laminate based on the fourth eigenvector.
2. The method according to claim 1, characterized in that The model structures of the deflection prediction neural network model and the thickness prediction neural network model both include an input layer, a first hidden layer, a second hidden layer and an output layer connected in sequence.
3. The method according to claim 2, characterized in that The activation functions used in the first hidden layer and the second hidden layer of the deflection prediction neural network model and the thickness prediction neural network model are Among them, x is the input value of the activation function, f(x) is the output value of the activation function, α is a hyperparameter, and the value range of α is 0 to 1.
4. The method according to claim 3, characterized in that The activation function used in the output layer is: Among them, f2(x1) is the output value of the activation function of the output layer, x1 is the input value of the activation function used by the output layer, and β is a hyperparameter.
5. The method according to claim 4, characterized in that The loss functions of the deflection prediction neural network model and the thickness prediction neural network model are both: Among them, Loss log -cosh is the prediction error, is the i1th true value, is the i1th prediction value, N is the number of training samples, i1 is the sample number, and cosh is the hyperbolic cosine function.
6. The method according to claim 5, characterized in that The parameter update formulas of the deflection prediction neural network model and the thickness prediction neural network model are both: Among them, Δω ij ω ij The correction amount, η is the learning rate; Among them, the number of iterations of training the deflection prediction neural network model and the thickness prediction neural network model are both n.
7. A device for obtaining an equivalent relationship between FRP laminates, characterized in that: The device comprises: A feature construction module: configured to use experimental data of multiple types of FRP laminates as sample data; select training data from the sample data; extract the FRP laminate thickness TA, geometric features, material properties and explosive loading conditions of the first type of FRP laminate to be equivalent to the first type of FRP laminate from the training data to construct a first feature vector; extract the maximum deflection, geometric features, material properties and explosive loading conditions of the second type of FRP laminate to be equivalent to the first type of FRP laminate from the training data to construct a second feature vector; A first training module: configured to train a deflection prediction neural network model based on the first eigenvector to obtain a trained deflection prediction neural network model; the deflection prediction neural network model takes the first eigenvector as input and determines the maximum deflection W of the first type of FRP laminate during the explosion response process based on the first eigenvector; A second training module: configured to train a thickness prediction neural network model based on the second eigenvector to obtain a trained thickness prediction neural network model; the thickness prediction neural network model uses the second eigenvector as input to generate a thickness TB of the second type FRP laminate corresponding to the maximum deflection of the second type FRP laminate; A series connection module: configured to sequentially connect the trained deflection prediction neural network model and the trained thickness prediction neural network model in series to serve as a trained series connection neural network model; Equivalent module: configured to obtain the FRP laminate thickness T of the first FRP laminate to be obtained for the equivalent relationship A1 , geometric features, material properties and explosive loading conditions, construct a third eigenvector; obtain the geometric features, material properties and explosive loading conditions of the second FRP laminate whose equivalent relationship is to be obtained; the trained series neural network model generates the maximum deflection W corresponding to the first FRP laminate based on the third eigenvector A1 , based on the maximum deflection W A1 , constructing a fourth eigenvector based on the geometric features, material properties and explosion loading conditions of the second FRP laminate to obtain the equivalent relationship, and generating the thickness of the second FRP laminate equivalent to the first FRP laminate based on the fourth eigenvector.
8. An electronic device, characterized in that: The device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 6.
9. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to cause the computer to execute the method according to any one of claims 1 to 6.